Variability, the problem of likeness, and the role of averages in face recognition
In the past, faces have been thought of as a single category of visual stimulus. Recent advancements in theory suggest we should split ‘faces’ into at least two subcategories - familiar and unfamiliar. With familiar people, we are able to tolerate a great deal of variability between different images of the same person. For unfamiliar people, though, this variability can harm our ability to recognise the same person in different photos. In this talk, I will discuss the role of variability in face learning, the mandatory nature of the formation of first impressions from faces, and the concept of an image of a familiar person being a ‘good likeness’. I will also discuss possible ways in which we might represent familiar people. One candidate representation is a face average, and I will discuss the use of averages for human and computer face recognition. Taken together, the issues of variability, likeness, and face averages have implications for photo-ID and face recognition in security settings.